Examples with ReducingStateDescriptor org.apache.flink.api.common.state.ReducingStateDescriptor used on opensource projects

Example 11 with ReducingStateDescriptor

use of org.apache.flink.api.common.state.ReducingStateDescriptor in project flink by apache.

the class WindowOperatorTest method testCleanupTimeOverflow.

@Test
public void testCleanupTimeOverflow() throws Exception {
    final int windowSize = 1000;
    final long lateness = 2000;
    ReducingStateDescriptor<Tuple2<String, Integer>> stateDesc = new ReducingStateDescriptor<>("window-contents", new SumReducer(), STRING_INT_TUPLE.createSerializer(new ExecutionConfig()));
    TumblingEventTimeWindows windowAssigner = TumblingEventTimeWindows.of(Time.milliseconds(windowSize));
    final WindowOperator<String, Tuple2<String, Integer>, Tuple2<String, Integer>, Tuple2<String, Integer>, TimeWindow> operator = new WindowOperator<>(windowAssigner, new TimeWindow.Serializer(), new TupleKeySelector(), BasicTypeInfo.STRING_TYPE_INFO.createSerializer(new ExecutionConfig()), stateDesc, new InternalSingleValueWindowFunction<>(new PassThroughWindowFunction<String, TimeWindow, Tuple2<String, Integer>>()), EventTimeTrigger.create(), lateness, null);
    OneInputStreamOperatorTestHarness<Tuple2<String, Integer>, Tuple2<String, Integer>> testHarness = createTestHarness(operator);
    testHarness.open();
    ConcurrentLinkedQueue<Object> expected = new ConcurrentLinkedQueue<>();
    long timestamp = Long.MAX_VALUE - 1750;
    Collection<TimeWindow> windows = windowAssigner.assignWindows(new Tuple2<>("key2", 1), timestamp, new WindowAssigner.WindowAssignerContext() {

        @Override
        public long getCurrentProcessingTime() {
            return operator.windowAssignerContext.getCurrentProcessingTime();
        }
    });
    TimeWindow window = Iterables.getOnlyElement(windows);
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), timestamp));
    // the garbage collection timer would wrap-around
    Assert.assertTrue(window.maxTimestamp() + lateness < window.maxTimestamp());
    // and it would prematurely fire with watermark (Long.MAX_VALUE - 1500)
    Assert.assertTrue(window.maxTimestamp() + lateness < Long.MAX_VALUE - 1500);
    // if we don't correctly prevent wrap-around in the garbage collection
    // timers this watermark will clean our window state for the just-added
    // element/window
    testHarness.processWatermark(new Watermark(Long.MAX_VALUE - 1500));
    // this watermark is before the end timestamp of our only window
    Assert.assertTrue(Long.MAX_VALUE - 1500 < window.maxTimestamp());
    Assert.assertTrue(window.maxTimestamp() < Long.MAX_VALUE);
    // push in a watermark that will trigger computation of our window
    testHarness.processWatermark(new Watermark(window.maxTimestamp()));
    expected.add(new Watermark(Long.MAX_VALUE - 1500));
    expected.add(new StreamRecord<>(new Tuple2<>("key2", 1), window.maxTimestamp()));
    expected.add(new Watermark(window.maxTimestamp()));
    TestHarnessUtil.assertOutputEqualsSorted("Output was not correct.", expected, testHarness.getOutput(), new Tuple2ResultSortComparator());
    testHarness.close();
}

Example 12 with ReducingStateDescriptor

use of org.apache.flink.api.common.state.ReducingStateDescriptor in project flink by apache.

the class WindowOperatorTest method testSideOutputDueToLatenessSessionZeroLatenessPurgingTrigger.

@Test
public void testSideOutputDueToLatenessSessionZeroLatenessPurgingTrigger() throws Exception {
    final int gapSize = 3;
    final long lateness = 0;
    ReducingStateDescriptor<Tuple2<String, Integer>> stateDesc = new ReducingStateDescriptor<>("window-contents", new SumReducer(), STRING_INT_TUPLE.createSerializer(new ExecutionConfig()));
    WindowOperator<String, Tuple2<String, Integer>, Tuple2<String, Integer>, Tuple3<String, Long, Long>, TimeWindow> operator = new WindowOperator<>(EventTimeSessionWindows.withGap(Time.seconds(gapSize)), new TimeWindow.Serializer(), new TupleKeySelector(), BasicTypeInfo.STRING_TYPE_INFO.createSerializer(new ExecutionConfig()), stateDesc, new InternalSingleValueWindowFunction<>(new ReducedSessionWindowFunction()), PurgingTrigger.of(EventTimeTrigger.create()), lateness, lateOutputTag);
    OneInputStreamOperatorTestHarness<Tuple2<String, Integer>, Tuple3<String, Long, Long>> testHarness = createTestHarness(operator);
    testHarness.open();
    ConcurrentLinkedQueue<Object> expected = new ConcurrentLinkedQueue<>();
    ConcurrentLinkedQueue<Object> sideExpected = new ConcurrentLinkedQueue<>();
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 1000));
    testHarness.processWatermark(new Watermark(1999));
    expected.add(new Watermark(1999));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 2000));
    testHarness.processWatermark(new Watermark(4998));
    expected.add(new Watermark(4998));
    // this will not be dropped because the session we're adding two has maxTimestamp
    // after the current watermark
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 4500));
    // new session
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 8500));
    testHarness.processWatermark(new Watermark(7400));
    expected.add(new Watermark(7400));
    // this will merge the two sessions into one
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 7000));
    testHarness.processWatermark(new Watermark(11501));
    expected.add(new StreamRecord<>(new Tuple3<>("key2-5", 1000L, 11500L), 11499));
    expected.add(new Watermark(11501));
    // new session
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 11600));
    testHarness.processWatermark(new Watermark(14600));
    expected.add(new StreamRecord<>(new Tuple3<>("key2-1", 11600L, 14600L), 14599));
    expected.add(new Watermark(14600));
    // this is side output as late
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 10000));
    sideExpected.add(new StreamRecord<>(new Tuple2<>("key2", 1), 10000));
    // this is also side output as late (we test that they are not accidentally merged)
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 10100));
    sideExpected.add(new StreamRecord<>(new Tuple2<>("key2", 1), 10100));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 14500));
    testHarness.processWatermark(new Watermark(20000));
    expected.add(new StreamRecord<>(new Tuple3<>("key2-1", 14500L, 17500L), 17499));
    expected.add(new Watermark(20000));
    testHarness.processWatermark(new Watermark(100000));
    expected.add(new Watermark(100000));
    ConcurrentLinkedQueue<Object> actual = testHarness.getOutput();
    ConcurrentLinkedQueue<StreamRecord<Tuple2<String, Integer>>> sideActual = testHarness.getSideOutput(lateOutputTag);
    TestHarnessUtil.assertOutputEqualsSorted("Output was not correct.", expected, actual, new Tuple2ResultSortComparator());
    TestHarnessUtil.assertOutputEqualsSorted("SideOutput was not correct.", sideExpected, (Iterable) sideActual, new Tuple2ResultSortComparator());
    testHarness.close();
}

Example 13 with ReducingStateDescriptor

use of org.apache.flink.api.common.state.ReducingStateDescriptor in project flink by apache.

the class WindowOperatorTest method testSideOutputDueToLatenessSessionZeroLateness.

@Test
public void testSideOutputDueToLatenessSessionZeroLateness() throws Exception {
    final int gapSize = 3;
    final long lateness = 0;
    ReducingStateDescriptor<Tuple2<String, Integer>> stateDesc = new ReducingStateDescriptor<>("window-contents", new SumReducer(), STRING_INT_TUPLE.createSerializer(new ExecutionConfig()));
    WindowOperator<String, Tuple2<String, Integer>, Tuple2<String, Integer>, Tuple3<String, Long, Long>, TimeWindow> operator = new WindowOperator<>(EventTimeSessionWindows.withGap(Time.seconds(gapSize)), new TimeWindow.Serializer(), new TupleKeySelector(), BasicTypeInfo.STRING_TYPE_INFO.createSerializer(new ExecutionConfig()), stateDesc, new InternalSingleValueWindowFunction<>(new ReducedSessionWindowFunction()), EventTimeTrigger.create(), lateness, lateOutputTag);
    OneInputStreamOperatorTestHarness<Tuple2<String, Integer>, Tuple3<String, Long, Long>> testHarness = createTestHarness(operator);
    testHarness.open();
    ConcurrentLinkedQueue<Object> expected = new ConcurrentLinkedQueue<>();
    ConcurrentLinkedQueue<Object> sideExpected = new ConcurrentLinkedQueue<>();
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 1000));
    testHarness.processWatermark(new Watermark(1999));
    expected.add(new Watermark(1999));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 2000));
    testHarness.processWatermark(new Watermark(4998));
    expected.add(new Watermark(4998));
    // this will not be dropped because the session we're adding two has maxTimestamp
    // after the current watermark
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 4500));
    // new session
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 8500));
    testHarness.processWatermark(new Watermark(7400));
    expected.add(new Watermark(7400));
    // this will merge the two sessions into one
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 7000));
    testHarness.processWatermark(new Watermark(11501));
    expected.add(new StreamRecord<>(new Tuple3<>("key2-5", 1000L, 11500L), 11499));
    expected.add(new Watermark(11501));
    // new session
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 11600));
    testHarness.processWatermark(new Watermark(14600));
    expected.add(new StreamRecord<>(new Tuple3<>("key2-1", 11600L, 14600L), 14599));
    expected.add(new Watermark(14600));
    // this is sideoutput as late, reuse last timestamp
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 10000));
    sideExpected.add(new StreamRecord<>(new Tuple2<>("key2", 1), 10000));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 14500));
    testHarness.processWatermark(new Watermark(20000));
    expected.add(new StreamRecord<>(new Tuple3<>("key2-1", 14500L, 17500L), 17499));
    expected.add(new Watermark(20000));
    testHarness.processWatermark(new Watermark(100000));
    expected.add(new Watermark(100000));
    ConcurrentLinkedQueue<Object> actual = testHarness.getOutput();
    ConcurrentLinkedQueue<StreamRecord<Tuple2<String, Integer>>> sideActual = testHarness.getSideOutput(lateOutputTag);
    TestHarnessUtil.assertOutputEqualsSorted("Output was not correct.", expected, actual, new Tuple2ResultSortComparator());
    TestHarnessUtil.assertOutputEqualsSorted("SideOutput was not correct.", sideExpected, (Iterable) sideActual, new Tuple2ResultSortComparator());
    testHarness.close();
}

Example 14 with ReducingStateDescriptor

use of org.apache.flink.api.common.state.ReducingStateDescriptor in project flink by apache.

the class TimeWindowTranslationTest method testReduceEventTimeWindows.

@Test
@SuppressWarnings("rawtypes")
public void testReduceEventTimeWindows() throws Exception {
    StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
    env.setStreamTimeCharacteristic(TimeCharacteristic.IngestionTime);
    DataStream<Tuple2<String, Integer>> source = env.fromElements(Tuple2.of("hello", 1), Tuple2.of("hello", 2));
    DataStream<Tuple2<String, Integer>> window1 = source.keyBy(0).timeWindow(Time.of(1000, TimeUnit.MILLISECONDS), Time.of(100, TimeUnit.MILLISECONDS)).reduce(new DummyReducer());
    OneInputTransformation<Tuple2<String, Integer>, Tuple2<String, Integer>> transform1 = (OneInputTransformation<Tuple2<String, Integer>, Tuple2<String, Integer>>) window1.getTransformation();
    OneInputStreamOperator<Tuple2<String, Integer>, Tuple2<String, Integer>> operator1 = transform1.getOperator();
    Assert.assertTrue(operator1 instanceof WindowOperator);
    WindowOperator winOperator1 = (WindowOperator) operator1;
    Assert.assertTrue(winOperator1.getTrigger() instanceof EventTimeTrigger);
    Assert.assertTrue(winOperator1.getWindowAssigner() instanceof SlidingEventTimeWindows);
    Assert.assertTrue(winOperator1.getStateDescriptor() instanceof ReducingStateDescriptor);
}

Example 15 with ReducingStateDescriptor

use of org.apache.flink.api.common.state.ReducingStateDescriptor in project flink by apache.

the class WindowOperatorTest method testReduceSessionWindows.

@Test
@SuppressWarnings("unchecked")
public void testReduceSessionWindows() throws Exception {
    closeCalled.set(0);
    final int sessionSize = 3;
    ReducingStateDescriptor<Tuple2<String, Integer>> stateDesc = new ReducingStateDescriptor<>("window-contents", new SumReducer(), STRING_INT_TUPLE.createSerializer(new ExecutionConfig()));
    WindowOperator<String, Tuple2<String, Integer>, Tuple2<String, Integer>, Tuple3<String, Long, Long>, TimeWindow> operator = new WindowOperator<>(EventTimeSessionWindows.withGap(Time.seconds(sessionSize)), new TimeWindow.Serializer(), new TupleKeySelector(), BasicTypeInfo.STRING_TYPE_INFO.createSerializer(new ExecutionConfig()), stateDesc, new InternalSingleValueWindowFunction<>(new ReducedSessionWindowFunction()), EventTimeTrigger.create(), 0, null);
    OneInputStreamOperatorTestHarness<Tuple2<String, Integer>, Tuple3<String, Long, Long>> testHarness = createTestHarness(operator);
    ConcurrentLinkedQueue<Object> expectedOutput = new ConcurrentLinkedQueue<>();
    testHarness.open();
    // add elements out-of-order
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 1), 0));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 2), 1000));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 3), 2500));
    // do a snapshot, close and restore again
    OperatorSubtaskState snapshot = testHarness.snapshot(0L, 0L);
    testHarness.close();
    testHarness = createTestHarness(operator);
    testHarness.setup();
    testHarness.initializeState(snapshot);
    testHarness.open();
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key1", 1), 10));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key1", 2), 1000));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key1", 3), 2500));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 4), 5501));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 5), 6000));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 5), 6000));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 6), 6050));
    testHarness.processWatermark(new Watermark(12000));
    expectedOutput.add(new StreamRecord<>(new Tuple3<>("key1-6", 10L, 5500L), 5499));
    expectedOutput.add(new StreamRecord<>(new Tuple3<>("key2-6", 0L, 5500L), 5499));
    expectedOutput.add(new StreamRecord<>(new Tuple3<>("key2-20", 5501L, 9050L), 9049));
    expectedOutput.add(new Watermark(12000));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 10), 15000));
    testHarness.processElement(new StreamRecord<>(new Tuple2<>("key2", 20), 15000));
    testHarness.processWatermark(new Watermark(17999));
    expectedOutput.add(new StreamRecord<>(new Tuple3<>("key2-30", 15000L, 18000L), 17999));
    expectedOutput.add(new Watermark(17999));
    TestHarnessUtil.assertOutputEqualsSorted("Output was not correct.", expectedOutput, testHarness.getOutput(), new Tuple3ResultSortComparator());
    testHarness.close();
}